改进动态图神经网络及其在三维牙齿模型分割中的应用

打开文本图片集

关键词：三维点云分割；动态图神经网络；口内扫描；牙齿实例分割

中图分类号：TP319 文献标志码：A 文章编号：1003-5168（2025）20-0019-05

DOI： 10.19968/j.cnki.hnkj.1003-5168.2025.20.004

Improvement of Dynamic Graph Neural Network and Its Application in 3D Tooth Model Segmentation

ZHAO Kai1LI Na²HE WenboHAN Huibo1 （1.Schoolof Computer Science and Information Engineering，Anyang Institute of Technology，Anyang ， China; 2.Department of Stomatology， Henan Vocational College of Nursing，Anyang 455Ooo， China）

Abstract： [Purposes] This study is to propose an improved dynamic graph neural network model for the problem of insufficient generalization of three-dimensional point cloud segmentation in the three一 dimensional model segmentation of intraoral teeth.[Methods] This paper conducts an in-depth analysis of the Dynamic Graph Convolutional Neural Network （DGCNN），clarifying that its original architecture is suitable for 3D point cloud classification and part segmentation.By improving the part segmentation branch of DGCNN， we propose a Dynamic Graph Neural Network Instance Segmentation network （DGISeg）for instance segmentation of intraoral tooth scan point cloud models.[Findings] The performance of the DGISeg model was then tested using the Teeth3DS dataset.The test results demonstrate that the DGISeg model achieves superior instance segmentation performance on the dataset compared to the PointNet model. [Conclusions] DGISeg retains the advantages of the DGCNN network while enhancing its instance perception capability through the proposed improvements，making it highly suitable for tooth instance segmentation following intraoral scanning.

Keywords： 3D point cloud segmentation; dynamic graph neural network; intraoral scanning; tooth instance segmentation

0 引言

三维点云分割是计算机视觉与几何处理的核心任务，旨在对点云或三维模型中的物体或部件进行细粒度语义划分与实例区分。（剩余5200字）